Google’s SynthID Bio can watermark AI-designed protein binders without breaking them
DeepMind's SynthID Bio watermarks AI-designed proteins for screening without hurting binder function.
Google DeepMind introduced SynthID Bio, a watermark for AI-designed proteins that places a signature in the amino-acid sequence and predicted 3D coordinates and can be checked after synthesis. Wet-lab tests of binders for VEGF-A, the SARS-CoV-2 spike receptor-binding domain, and PD-L1 found watermarked designs matched unwatermarked ones on hit rate, affinity, and sequence diversity. DeepMind positions the mark as a signal for DNA-synthesis screening and database submissions, while noting resistance to deliberate tampering is still unsolved. Early tests watermarking an Evo 2-designed bacteriophage genome found the phages remained functional, and the paper, code, in-vitro data, and weights are being released.
- SynthID Bio hides a checkable signature in designed protein sequences and structures.
- Watermarked binders matched unwatermarked designs on three wet-lab targets.
- DeepMind aims the mark at DNA-synthesis screening of unfamiliar AI designs.
- Early Evo 2 bacteriophage tests found watermarked phages still functional.
- Tamper resistance is unsolved; code, data, and weights are being released.
- Vendors
From the original article
2 min read at helpnetsecurity.com
Google DeepMind has built SynthID Bio, a watermark for AI-designed proteins, and shown in wet-lab tests on protein binders that it leaves their function intact. The method hides a signature in the amino acid sequence of a designed protein and in the atomic coordinates of a predicted 3D structure. DeepMind says the signature can be checked on the physical protein after synthesis.
By Anamarija Pogorelec · helpnetsecurity.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.